Re: Build failures in -next due to 'linux/io.h: Add pci_remap_cfgspace() interface'

[Lorenzo Pieralisi]

On Mon, Apr 24, 2017 at 02:18:09PM -0500, Bjorn Helgaas wrote:
> On Mon, Apr 24, 2017 at 3:56 AM, Lorenzo Pieralisi
> <lorenzo.pieralisi@xxxxxxx> wrote:
> > On Sun, Apr 23, 2017 at 05:10:55PM -0700, Guenter Roeck wrote:
> >> Hi,
> >>
> >> -next fails to build for several architectures due to commit 'linux/io.h: Add
> >> pci_remap_cfgspace() interface'. Error message is
> >>
> >> include/linux/io.h: In function âpci_remap_cfgspaceâ:
> >> include/linux/io.h:108:9: error: implicit declaration of function âioremap_nocacheâ
> >>
> >> Affected architectures (not necessarily a complete list):
> >>
> >> um
> >> score
> >> s390
> >>
> >> Bisect results for s390 and score are below. I did not bother to bisect
> >> the um build; the error message is the same.
> >
> > Sorry about that, kbuild did not catch those. Patch below should fix it
> > (tested it on um/s390); either I do that or I will have to remove the
> > static inline and make it a:
> >
> > #define pci_remap_cfgspace ioremap_nocache
> >
> > It is nice to see how consistent the ioremap interface is across arches :)
> >
> > Bjorn please let me know how you prefer fixing it, I will post a patch
> > accordingly.
> >
> > Lorenzo
> >
> > -- >8 --
> > diff --git a/include/linux/io.h b/include/linux/io.h
> > index 3934aba..2195d9e 100644
> > --- a/include/linux/io.h
> > +++ b/include/linux/io.h
> > @@ -90,6 +90,7 @@ void *devm_memremap(struct device *dev, resource_size_t offset,
> >
> >  void *__devm_memremap_pages(struct device *dev, struct resource *res);
> >
> > +#ifdef CONFIG_PCI
> >  /*
> >   * The PCI specifications (Rev 3.0, 3.2.5 "Transaction Ordering and
> >   * Posting") mandate non-posted configuration transactions. There is
> > @@ -108,6 +109,7 @@ static inline void __iomem *pci_remap_cfgspace(phys_addr_t offset,
> >         return ioremap_nocache(offset, size);
> >  }
> >  #endif
> > +#endif
> 
> I folded this #ifdef wrapper into the patch and updated my "next"
> branch.  Hopefully that's enough?

It is not ideal but it should be enough to the best of my knowledge,
I will keep testing it.

I am not sure there is a cleaner way to provide a default implementation
unless we resort to #define pci_remap_cfgspace ioremap_nocache instead
of the static inline version.

Thanks,
Lorenzo